Plansmith
Live demos

Construction management system demo: 4 live apps built by AI models

Every app below is a working construction management system, built by an AI coding model from the PlanSmith Construction Management planner — running, with sample data, not a mock-up. Open one, sign in with the demo account it shows, and try it.

Built by Claude Fable 5, GPT-5.6 Sol, Grok 4.6 and Gemini 3.7 FlashAudited ·
The apps

Open one and try it.

Confidentiality as a data model — live demoLive
Built by Claude Fable 5

Confidentiality as a data model

“ForgeLedger” — the highest-scoring build of the four, and the one that treats a jobsite as what it is: parties who collaborate and compete at the same time. It keeps a share register, so each document, drawing and report is granted to a named party and everything else is refused. All three subcontractors sit on the same project, so the isolation cannot fall out of project membership — tested in both directions it leaked nothing, and of 416 files the plumbing sub could read exactly the one behind her grant. Pick a role on /login to get straight in — project manager, site supervisor, subcontractor or client. Those buttons are off by default in the build and switched on deliberately here, and they run the same session path as a password sign-in rather than a shortcut around it. You land on its public project showcase; sign in from there. By hand works too: dana.whitfield@northstarridge.example / pm-demo-2026.

Open live demo
The one that works with JavaScript off — live demoLive
Built by GPT-5.6 Sol

The one that works with JavaScript off

“ForgeLedger” again, from the same planner — worth opening beside the others. Its sign-in is a real form rather than a script, so it completes with JavaScript entirely disabled and no credential ever reaches the address bar. Its permissions are a gradient rather than a switch: the project manager sees four tasks, the supervisor three, the subcontractor exactly the one that is theirs, and cost, estimating and procurement are closed to the subcontractor and the owner alike. The highest-sounding role is deliberately not a superset — the project manager is refused the owner’s records screen. Accounts are default-deny: with no password configured for a role, that account does not exist at all. Pick a role on /login. This build ships no demo shortcut of its own — its sign-in page says so — so the buttons here post a seeded credential to the same endpoint the password form uses rather than routing around it. You land on its public bid board — a planner feature in its own right — and sign in from there. By hand works too: pm@northstarridge.example / ForgeLedgerDemo2026.

Open live demo
Two companies, to prove they stay apart — live demoLive
Built by Grok 4.6

Two companies, to prove they stay apart

“ForgeLedger” — the leanest build in the whole demo fleet: one production dependency, no framework, a single server file, and nothing for a future advisory to land on. It is also the only construction build that seeds a second construction company on purpose so customer separation can be demonstrated rather than asserted, and that separation held on every route we drove. Its first audit found demo buttons that handed any visitor a project-manager session with no way to switch them off, and a verification suite covering step one of sixteen with fifteen checks never written; both were repaired and re-verified, and it was rescored 84 → 93. You land on its public work board; sign in from there and pick a role. The demo buttons are switched on deliberately here and off by default everywhere else.

Open live demo
The richest data model, after repair — live demoLive
Built by Gemini 3.7 Flash

The richest data model, after repair

“ForgeLedger” — the fourth build from the same planner, and the one that models the work most thoroughly: WBS codes with real float and critical-path flags, CSI cost codes connecting an estimate to a commitment to an actual, sheet numbers with disciplines and revisions. It arrived with the vertical’s defining defect: of forty-six role checks, forty-five protected a write and exactly one protected a read, so a trade subcontractor read every rival’s contract value — and the one table it did protect leaked anyway through a CSV export with no role check. Two repair rounds closed it: rival subcontractors now see only their own purchase orders with zero overlap, the export returns exactly the row set the screen does, and the public page no longer publishes internal reviewer notes. What is still open is the owner, who reads the contractor’s commitments, estimating build-up and budget in full. Pick a role from the sandbox switcher to get in.

Open live demo
How they scored

Same planner, different results.

RankBuildModelScoreIn one line
1ForgeLedgerClaude Fable 596The one to copy. Nothing we tried to break came apart.
2ForgeLedgerGPT-5.6 Sol92Sign-in survives with JavaScript switched off. Almost nothing else does.
3ForgeLedgerGrok 4.684Strongest separation between customers, and the leanest thing here by far.
4ForgeLedgerGemini 3.7 Flash74The richest domain model here, with the confidentiality left out of it.

Scores are as found at audit. The full findings for every build — what it does well and what we found — are on the construction management build-off board.